arxiv: 2604.19450 · v1 · submitted 2026-04-21 · 💻 cs.CG · math.CO

Recognition: unknown

Local Depth-Based Corrections to Maxmin Landmark Selection for Lazy Witness Persistence

Yifan Zhang

Pith reviewed 2026-05-10 01:12 UTC · model grok-4.3

classification 💻 cs.CG math.CO

keywords lazy witness persistencemaxmin landmarksdepth-based correctionspersistent homologytopological data analysislandmark selectioncomputational geometry

0 comments

The pith

Support-weighted partial recentering of maxmin seeds improves geometric quality in lazy witness persistence without changing the thresholded H1 summary

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops local depth-based corrections to maxmin landmark selection for lazy witness persistence. It begins with standard maxmin seeds, partitions the point cloud into nearest-seed cells, and then moves or replaces each seed toward a deep representative of its cell, scaling the movement by cell support size. Mathematical proofs establish local guarantees including convex-core robustness from halfspace depth and 2r cover bounds. Experiments on planar synthetic data and silhouettes show consistent geometric improvement while the thresholded H1 persistence summary remains the same as the maxmin baseline, with mixed topological results in a modest 3D extension. This frames the work as a controlled study of a local refinement rather than a global approximation result.

Core claim

Starting from maxmin seeds, the support-weighted partial recentering rule partitions the cloud into nearest-seed cells and moves each seed toward a deep representative of its cell, with the movement scaled by cell support. This yields local geometric guarantees: a convex-core robustness lemma from halfspace depth, a 2r cover bound, and projected cover bounds. In the main planar experiments the corrected landmarks produce a consistent geometric improvement over plain maxmin while the thresholded H1 summary stays unchanged; the 3D torus test shows the same geometric tendency but only mixed topological behavior.

What carries the argument

Support-weighted partial recentering, which scales the displacement of each maxmin seed toward a high-depth representative of its nearest-seed cell by the relative support size of that cell.

If this is right

The corrected landmarks satisfy convex-core robustness and 2r cover bounds derived from depth measures.
Geometric error decreases consistently on planar synthetic benchmarks and MPEG-7 silhouettes.
The thresholded H1 persistence summary remains identical to the maxmin baseline in the planar regime.
In 3D torus tests the geometric improvement persists while topological stability holds only partially.
The method supplies a practical local alternative to maxmin rather than a global witness-approximation theorem.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same depth-weighted adjustment could be tested inside other landmark selectors or persistence constructions to see whether the geometric gain generalizes.
Applications that tolerate small changes in landmark placement might achieve equivalent topological output with fewer landmarks after recentering.
The mixed 3D topological results point to a need for controlled tests that vary noise level and cell-size distribution to locate the boundary between stable and unstable H1 behavior.
Combining the local recentering step with a final global optimization pass could tighten both geometry and topology simultaneously.

Load-bearing premise

The local depth-based adjustments preserve the essential topological features captured by the lazy witness complex across the tested regimes without introducing new artifacts not captured by the H1 threshold.

What would settle it

A point cloud where applying support-weighted partial recentering causes the thresholded H1 persistence diagram to gain or lose long-lived features compared with plain maxmin would falsify the topological preservation claim.

Figures

Figures reproduced from arXiv: 2604.19450 by Yifan Zhang.

**Figure 2.** Figure 2: Parameter sensitivity for support-weighted partial recentering. Over the tested [PITH_FULL_IMAGE:figures/full_fig_p018_2.png] view at source ↗

**Figure 3.** Figure 3: Aggregate view of the MPEG-7 silhouette-based case study. Support-weighted partial [PITH_FULL_IMAGE:figures/full_fig_p019_3.png] view at source ↗

read the original abstract

We study a family of local depth-based corrections to maxmin landmark selection for lazy witness persistence. Starting from maxmin seeds, we partition the cloud into nearest-seed cells and replace or move each seed toward a deep representative of its cell. The principal implemented variant, \emph{support-weighted partial recentering}, scales the amount of movement by cell support. The contributions are both mathematical and algorithmic. On the mathematical side, we prove local geometric guarantees for these corrections: a convex-core robustness lemma derived from halfspace depth, a $2r$ cover bound for subset recentering, and projected cover bounds for the implemented partial-recentering rules. On the algorithmic side, we identify a practically effective variant through a layered empirical study consisting of planar synthetic benchmarks, a parameter-sensitivity study, and an MPEG-7 silhouette benchmark, together with a modest three-dimensional torus extension. The main planar experiments show that support-weighted partial recentering gives a consistent geometric improvement over maxmin while preserving the thresholded $H_1$ summary used in the study. The three-dimensional experiment shows the same geometric tendency but only mixed topological behavior. The paper should therefore be read as a controlled study of a local depth-based alternative to maxmin, rather than as a global witness-approximation theorem or a claim of uniform empirical superiority.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper adds local geometric lemmas for depth-based tweaks to maxmin landmarks but leaves the key claim of H1 preservation empirical and dataset-specific.

read the letter

The main takeaway is that this work defines a family of depth-based corrections to maxmin landmark selection for lazy witness persistence, proves some local geometric properties for them, and shows through benchmarks that one variant improves geometry while keeping the thresholded H1 the same in planar cases. The support-weighted partial recentering stands out as the practical choice from their tests. What is actually new is the pairing of halfspace depth with recentering, plus the convex-core robustness lemma, the 2r cover bound for subset recentering, and the projected bounds for the partial rules. Those lemmas are independent of the experiments and give concrete local guarantees that were not combined this way before. The layered empirical section on synthetic planar data, parameter sensitivity, MPEG-7 silhouettes, and a 3D torus extension is also a plus; it is controlled and reports consistent geometric gains in 2D. The soft spot is the gap between the proven local bounds and the topological claim. There is no theorem connecting the cover bounds to stability of the witness persistence diagram, so the assertion that no new artifacts appear outside the H1 threshold rests on the specific datasets and thresholding used. The 3D torus results are mixed, which weakens the preservation argument, and the experiments lack error bars or statistical tests, making the consistency harder to assess. This is a narrow but honest study for people already working on witness complexes and landmark selection in TDA. A reader who needs better local approximations for shape analysis or geometry could try the variants on their own data. It has enough original math and careful experiments to deserve a serious referee, though the authors should be asked to either strengthen the geometry-to-topology link or qualify the H1 observations more clearly as empirical. I would send it for peer review.

Referee Report

2 major / 3 minor

Summary. The paper studies local depth-based corrections to maxmin landmark selection for lazy witness persistence. Starting from maxmin seeds, it partitions the point cloud into nearest-seed cells and applies corrections (replace or move toward a deep representative), with the principal variant being support-weighted partial recentering. Mathematical contributions include proofs of a convex-core robustness lemma from halfspace depth, a 2r cover bound for subset recentering, and projected cover bounds for partial recentering. Empirically, planar synthetic and MPEG-7 benchmarks show consistent geometric improvement over maxmin with preserved thresholded H1, while a 3D torus extension shows similar geometric trends but mixed topological results. The work is framed as a controlled study of the alternative rather than a general approximation theorem.

Significance. If the empirical observations on H1 preservation generalize, the approach could provide a practical, locally justified improvement to landmark selection in persistent homology pipelines, yielding better geometric fidelity in the witness complex without altering the key topological summaries. The local geometric proofs are a clear strength, offering rigorous insight into the behavior of the corrections independent of the empirical fits.

major comments (2)

[empirical study sections] Planar experiments and 3D torus extension: the claim of consistent geometric improvement with H1 preservation is presented without error bars, multiple random seeds, or statistical tests on the reported gains; this makes the 'consistent' qualifier difficult to assess quantitatively, especially given the mixed 3D topological outcomes.
[mathematical guarantees and abstract] Mathematical contributions and central claim: the proven local cover bounds and convex-core robustness are independent of the empirical fits, but no theorem or argument connects these bounds to the stability of the lazy witness complex persistence diagram; the assertion that no new artifacts appear outside the H1 threshold therefore rests entirely on dataset-specific thresholding rather than a general guarantee.

minor comments (3)

[abstract] The abstract and conclusion could more explicitly qualify the 3D mixed results as a limitation rather than a minor extension.
[algorithmic description] Notation for cell support and the scaling in support-weighted partial recentering should be defined more formally early in the algorithmic description to aid readability.
[figures] Figure captions for the recentering illustrations would benefit from explicit labels indicating the depth-based movement vectors and cell boundaries.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive report, which correctly identifies both the local geometric contributions and the scope limitations of the work. We respond to each major comment below and indicate the revisions we will make.

read point-by-point responses

Referee: [empirical study sections] Planar experiments and 3D torus extension: the claim of consistent geometric improvement with H1 preservation is presented without error bars, multiple random seeds, or statistical tests on the reported gains; this makes the 'consistent' qualifier difficult to assess quantitatively, especially given the mixed 3D topological outcomes.

Authors: We agree that the empirical sections would be strengthened by additional statistical controls. The current presentation relies on single runs of maxmin and reports point estimates without variability measures. In the revised manuscript we will execute the planar benchmarks across multiple random seeds for the initial maxmin selection, report mean geometric improvements together with standard deviations, and include simple statistical comparisons (e.g., paired t-tests) against the baseline. For the 3D torus extension we will expand the discussion to characterize the specific instances in which the thresholded H1 summary changes, thereby clarifying the mixed topological behavior. revision: yes
Referee: [mathematical guarantees and abstract] Mathematical contributions and central claim: the proven local cover bounds and convex-core robustness are independent of the empirical fits, but no theorem or argument connects these bounds to the stability of the lazy witness complex persistence diagram; the assertion that no new artifacts appear outside the H1 threshold therefore rests entirely on dataset-specific thresholding rather than a general guarantee.

Authors: The referee accurately observes that the mathematical results (convex-core robustness, 2r cover bound, and projected cover bounds) are purely local geometric statements and do not yield a stability guarantee for the persistence diagrams of the lazy witness complex. The manuscript already states that the work is a controlled study of a local correction rather than a general approximation theorem, and the H1 preservation claim is explicitly tied to the thresholding used in the reported benchmarks. We will revise the abstract and the final discussion paragraph to reinforce this framing and to remove any phrasing that could be read as implying a general absence of new artifacts. revision: partial

Circularity Check

0 steps flagged

No significant circularity; proofs and empirical claims remain independent

full rationale

The paper separates its mathematical contributions (convex-core robustness lemma from halfspace depth, 2r cover bound for subset recentering, and projected cover bounds for partial recentering) from its empirical study of support-weighted partial recentering on planar benchmarks and 3D torus. No equation or claim reduces a 'prediction' to a fitted parameter by construction, nor does any central result depend on a self-citation chain or ansatz smuggled from prior work. The H1 preservation statement is explicitly presented as an empirical observation rather than a derived theorem, avoiding any definitional loop. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Relies on standard properties of halfspace depth and maxmin selection; new lemmas are proved from these. No free parameters or invented entities are introduced in the abstract description.

axioms (1)

domain assumption Halfspace depth defines a robust central representative within each nearest-seed cell
Invoked to justify moving seeds toward deep points while preserving coverage

pith-pipeline@v0.9.0 · 5527 in / 1273 out tokens · 34397 ms · 2026-05-10T01:12:08.712892+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

27 extracted references · 16 canonical work pages

[1]

Topological estimation using witness complexes

Vin de Silva and Gunnar Carlsson. Topological estimation using witness complexes. In Proceedings of the Eurographics Symposium on Point-Based Graphics, pages 157–166, 2004. DOI:https://doi.org/10.2312/SPBG/SPBG04/157-166

work page doi:10.2312/spbg/spbg04/157-166 2004
[2]

Porter, Ulrike Tillmann, Peter Grindrod, and Heather A

Nina Otter, Mason A. Porter, Ulrike Tillmann, Peter Grindrod, and Heather A. Harrington. A roadmap for the computation of persistent homology.EPJ Data Science, 6(17):1–38,
[3]

DOI:https://doi.org/10.1140/epjds/s13688-017-0109-5

work page doi:10.1140/epjds/s13688-017-0109-5
[4]

Bernadette J. Stolz. Outlier-robust subsampling techniques for persistent homology.Journal of Machine Learning Research, 24(90):1–35, 2023. Available at:https://www.jmlr.org/ papers/v24/21-1526.html

2023
[5]

Topological data analysis with ϵ-net induced lazy witness complexes.arXiv preprint arXiv:1906.06122, 2019

Naheed Anjum Arafat, Debabrota Basu, and Stéphane Bressan. Topological data analysis with ϵ-net induced lazy witness complexes.arXiv preprint arXiv:1906.06122, 2019. DOI: https://doi.org/10.48550/arXiv.1906.06122. 23

work page doi:10.48550/arxiv.1906.06122 1906
[6]

Adaptive approximation of persistent homology.Journal of Applied and Computational Topology, 8:2327–2366, 2024

Maria Herick, Michael Joachim, and Jan Vahrenhold. Adaptive approximation of persistent homology.Journal of Applied and Computational Topology, 8:2327–2366, 2024. DOI: https://doi.org/10.1007/s41468-024-00192-7

work page doi:10.1007/s41468-024-00192-7 2024
[7]

Tverberg’s theorem is 50 years old: A survey.Bulletin of the American Mathematical Society, 55(4):459–492, 2018

Imre Bárány and Pablo Soberón. Tverberg’s theorem is 50 years old: A survey.Bulletin of the American Mathematical Society, 55(4):459–492, 2018. DOI:https://doi.org/10. 1090/bull/1634

2018
[8]

American Mathematical Society, 2010

Herbert Edelsbrunner and John Harer.Computational Topology: An Introduction. American Mathematical Society, 2010. DOI:https://doi.org/10.1090/mbk/069

work page doi:10.1090/mbk/069 2010
[9]

Topology and data.Bulletin of the American Mathematical Society, 46(2):255–308, 2009

Gunnar Carlsson. Topology and data.Bulletin of the American Mathematical Society, 46(2):255–308, 2009. DOI:https://doi.org/10.1090/S0273-0979-09-01249-X

work page doi:10.1090/s0273-0979-09-01249-x 2009
[10]

Guibas and Steve Y

Leonidas J. Guibas and Steve Y. Oudot. Reconstruction using witness complexes.Dis- crete & Computational Geometry, 40(3):325–356, 2008. DOI:https://doi.org/10.1007/ s00454-008-9094-6

2008
[11]

Alpha-beta witness complexes

Dominique Attali, Herbert Edelsbrunner, John Harer, and Yuriy Mileyko. Alpha-beta witness complexes. InAlgorithms and Data Structures, volume 4619 ofLecture Notes in Computer Science, pages 386–397. Springer, 2007. DOI:https://doi.org/10.1007/ 978-3-540-73951-7_34

2007
[12]

Bodlaender, Corinne Feremans, Alexander Grigoriev, Eelko Penninkx, Ren´ e Sitters, and Thomas Wolle

Tamal K. Dey, Fengtao Fan, and Yusu Wang. Graph induced complex on point data.Com- putational Geometry, 48(8):575–588, 2015. DOI:https://doi.org/10.1016/j.comgeo. 2015.04.003

work page doi:10.1016/j.comgeo 2015
[13]

Subsampling methods for persistent homology

Frédéric Chazal, Brittany Terese Fasy, Fabrizio Lecci, Bertrand Michel, Alessandro Rinaldo, and Larry Wasserman. Subsampling methods for persistent homology. InProceedings of the 32nd International Conference on Machine Learning, volume 37 ofProceedings of Machine Learning Research, pages 2143–2151. PMLR, 2015. Available at: https: //proceedings.mlr.press...

2015
[14]

Sampling from pseudometric spaces with applications to persistent homology.arXiv preprint arXiv:2212.09826, 2022

Sammy Brunson and Clara Skaf. Sampling from pseudometric spaces with applications to persistent homology.arXiv preprint arXiv:2212.09826, 2022. DOI:https://doi.org/10. 48550/arXiv.2212.09826

work page arXiv 2022
[15]

Computing a centerpoint of a finite planar set of points in linear time.Discrete & Computational Geometry, 12(3):291–312, 1994

Shreesh Jadhav and Asish Mukhopadhyay. Computing a centerpoint of a finite planar set of points in linear time.Discrete & Computational Geometry, 12(3):291–312, 1994. DOI: https://doi.org/10.1007/BF02574382

work page doi:10.1007/bf02574382 1994
[16]

Agarwal, Micha Sharir, and Emo Welzl

Pankaj K. Agarwal, Micha Sharir, and Emo Welzl. Algorithms for center and Tverberg points.ACM Transactions on Algorithms, 5(1):Article 5, 2008. DOI:https://doi.org/ 10.1145/1435375.1435380

work page doi:10.1145/1435375.1435380 2008
[17]

Approximating tverberg points in linear time for any fixed dimension.Discrete & Computational Geometry, 50(2):520–535, 2013

Wolfgang Mulzer and Daniel Werner. Approximating tverberg points in linear time for any fixed dimension.Discrete & Computational Geometry, 50(2):520–535, 2013. DOI: https://doi.org/10.1007/s00454-013-9528-7

work page doi:10.1007/s00454-013-9528-7 2013
[18]

Katharine Turner, Sayan Mukherjee, and Doug M. Boyer. Persistent homology transform for modeling shapes and surfaces.Information and Inference, 3(4):310–344, 2014. DOI: https://doi.org/10.1093/imaiai/iau011

work page doi:10.1093/imaiai/iau011 2014
[19]

Exploring uses of persistent homology for statistical analysis of landmark-based shape data.Journal of Multivariate Analysis, 101(9):2184–2199,

Jennifer Gamble and Giseon Heo. Exploring uses of persistent homology for statistical analysis of landmark-based shape data.Journal of Multivariate Analysis, 101(9):2184–2199,
[20]

DOI:https://doi.org/10.1016/j.jmva.2010.04.016. 24

work page doi:10.1016/j.jmva.2010.04.016 2010
[21]

Shape matching and object recognition using shape contexts.IEEE Transactions on Pattern Analysis and Machine Intelligence, 24(4):509–522, 2002

Serge Belongie, Jitendra Malik, and Jan Puzicha. Shape matching and object recognition using shape contexts.IEEE Transactions on Pattern Analysis and Machine Intelligence, 24(4):509–522, 2002. DOI:https://doi.org/10.1109/34.993558

work page doi:10.1109/34.993558 2002
[22]

Shape descriptors for non-rigid shapes with a single closed contour

Longin Jan Latecki, Rolf Lakämper, and Ulrich Eckhardt. Shape descriptors for non-rigid shapes with a single closed contour. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 424–429, 2000. DOI:https://doi.org/10.1109/ CVPR.2000.855850

work page arXiv 2000
[23]

The GUDHI library: Simplicial complexes and persistent homology

Clément Maria, Jean-Daniel Boissonnat, Marc Glisse, and Mariette Yvinec. The GUDHI library: Simplicial complexes and persistent homology. InInternational Congress on Mathematical Software, volume 8592 ofLecture Notes in Computer Science, pages 167–174. Springer, 2014. DOI:https://doi.org/10.1007/978-3-662-44199-2_28

work page doi:10.1007/978-3-662-44199-2_28 2014
[24]

GUDHI Editorial Board, 3.12.0 edition, 2026

The GUDHI Project.GUDHI User and Reference Manual. GUDHI Editorial Board, 3.12.0 edition, 2026. URL:https://gudhi.inria.fr/doc/3.12.0/

2026
[25]

Filtered complexes

Clément Maria. Filtered complexes. InGUDHI User and Reference Manual. GUDHI EditorialBoard, 3.12.0edition, 2026. URL:https://gudhi.inria.fr/doc/3.12.0/group_ _simplex__tree.html

2026
[26]

Persistent cohomology

Clément Maria. Persistent cohomology. InGUDHI User and Reference Manual. GUDHI EditorialBoard, 3.12.0edition, 2026. URL:https://gudhi.inria.fr/doc/3.12.0/group_ _persistent__cohomology.html

2026
[27]

Experimental datasets and supporting scripts for depth-based landmark recentering in lazy witness complexes, 2026

Yifan Zhang. Experimental datasets and supporting scripts for depth-based landmark recentering in lazy witness complexes, 2026. Dataset. DOI:https://doi.org/10.5281/ zenodo.19676292. 25

2026